RuntimeError: shape '[1, 1, 32, 32]' is invalid for input of size 993

[ReedOnePeck]

当文本输入比较长的时候会随机报这个错误，多跑几次有时候会报错，有时候就可以跑通。
File "/usr/local/lib/python3.12/dist-packages/torch/utils/_contextlib.py", line 116, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/diffusers/pipelines/glm_image/pipeline_glm_image.py", line 621, in call
prior_token_ids, prior_token_image_ids = self.generate_prior_tokens(
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/diffusers/pipelines/glm_image/pipeline_glm_image.py", line 314, in generate_prior_tokens
prior_token_ids = self._upsample_token_ids(prior_token_ids_d32, token_h, token_w)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.12/dist-packages/diffusers/pipelines/glm_image/pipeline_glm_image.py", line 256, in _upsample_token_ids
token_ids = token_ids.view(1, 1, token_h, token_w)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: shape '[1, 1, 32, 32]' is invalid for input of size 993

[D1-3105]

yeah i faced the same, the solution was:

        prior_token_ids_d32 = torch.nn.functional.pad(prior_token_ids_d32, (0, token_h*token_w - prior_token_ids_d32.shape[0]))
        prior_token_ids = self._upsample_token_ids(prior_token_ids_d32, token_h, token_w)